Presentation is loading. Please wait.

Presentation is loading. Please wait.

Authors: Duo Liu, Bei Hua, Xianghui Hu and Xinan Tang

Similar presentations


Presentation on theme: "Authors: Duo Liu, Bei Hua, Xianghui Hu and Xinan Tang"— Presentation transcript:

1 High-performance Packet Classification Algorithm for Many-core and Multithreaded Network Processor
Authors: Duo Liu, Bei Hua, Xianghui Hu and Xinan Tang Publisher: Proceedings of the 2006 international conference on Compilers, architecture and synthesis for embedded systems CASES '06 Present: Yu-Ru Chung Date: December

2 outline Method description (RFC and Bit-RFC) Design issues
Simulation and performance analysis

3

4 RFC

5 Bitmap RFC

6

7

8 Design issues Memory space reduction Instruction selection
Data allocation Task partitioning Latency hiding

9 Simulation and performance analysis experimental setup
The RFC algorithm needs six chunks (corresponding to 16 bits lower/higher src/dst IP address, 16 bits src/dst port number respectively) in Phase 0, two CPTs (CPT X and CPT Y) in Phase 1, and one CPT (CPT Z) in Phase 2. To compare with the RFC algorithm, we implemented 4-deminsional Bitmap-RFC packet classification with the same three phases. The compressed tables used by Bitmap-RFC are called CCPT X’/Y’/Z’. CL#1, CL#2, CL#3, and each has 1000, 2000, 3000 rules respectively. For the OC-192 core routers a minimal packet has 49 bytes (9-byte PPP header + 20-byte IPv4 header + 20-byte TCP header).

10

11 Memory Requirement of RFC and Bitmap-RFC
To find out the maximum number of rules that RFC and Bitmap-RFC can hold on an IXP2800, we use two metrics: • The minimum number of rules that causes one of tables lager than 64MBytes (the size of one SRAM controller). • The minimum number of rules that causes the total size of all tables larger than 256MBytes (the total size of SRAM).

12

13 Relative Speedups The reason of sub-linear speedup is partially caused by the saturation of command request FIFOs and SRAM buses.

14 Instruction Selection

15

16 Memory Impacts

17 Task Partitioning

18 Latency Hiding


Download ppt "Authors: Duo Liu, Bei Hua, Xianghui Hu and Xinan Tang"

Similar presentations


Ads by Google